Oilogosphere veteran Scott Reynolds popped up in the comments from my last post and made a good point. I’d posted a table showing the Corsi% for teams after they won defensive zone faceoffs last year:
Scott made the following (very salient) point:
I wonder how much of the difference between teams is talent and how much is luck in the current sample. It’s not a huge number of shot attempts. It would be interesting to distribute last season’s games randomly into two buckets (“A” games and “B” games) to see how much difference there is between the two. That cuts the overall sample in half, so that might not be the best method, but it seems possible to me that some bad teams are near the top of the list because the sample is too small to separate the wheat from the chaff.
I should point out that I tend to think that you’ll get signal with smaller samples when you’re doing something like this, as opposed to looking at games as a whole, because you’re looking at more of an apples to apples situation. In every case, we’re starting with a situation deemed a faceoff win by the scorer, which means possession of the puck in your own end.
To run a test along the lines that Scott’s suggesting, I went through each team’s shot attempts for/against following defensive zone wins in order, labeling 1 and then 2. I then looked at the relationship between the two. If it was all signal or, to put it differently, if half the sample for each team was big enough to be unaffected by randomness, we’d expect each team to have exactly the same Corsi% in Group 1 as it did in Group 2.
You should be able to tell from the image that it looks like there’s some sort of relationship between the Group 1 and Group 2 numbers. If Group 1 predicted Group 2 perfectly, it would look like a line. We don’t have that but we do have a correlation of 0.47, which isn’t terrible. It suggests that there’s a reasonable amount of signal here.
You can see that, with 27 teams, Group 1 is within about +/- 8 points of Group 2. Three teams are off on their own. New Jersey’s fairly easy to understand: the Group 1 and Group 2 samples are much smaller, which is going to allow for more noise. I think this problem is fairly sample size driven – Tampa has the third smallest Group 1/2 and St. Louis is fourth smallest. Everyone else is within the threshold.
So, to Scott’s point, it’s obviously a possibility, although the fact that we’re looking at full season samples rather than half season samples reduces the likelihood that that’s the issue. That being said, of the bad to average teams that I highlighted who posted a Corsi% of 40%+ after a DZW, there may well be one or two who just got lucky. On balance though, I’m inclined to believe that they were, in fact, better than the Oilers at doing this, whether due to talent or tactics.Email Tyler Dellow at email@example.com